Single-pixel ensemble correlation algorithm for boundary measurement on axisymmetric boattail surface

Journal of Science and Technique - N.208 (6-2020) - Le Quy Don Technical University 89 SINGLE-PIXEL ENSEMBLE CORRELATION ALGORITHM FOR BOUNDARY MEASUREMENT ON AXISYMMETRIC BOATTAIL SURFACE Tran The Hung* Le Quy Don Technical University Abstract Particle image velocimetry (PIV) measurement is an important technique in analyzing velocity fields. However, in traditional cross-correlation algorithm, the resolution of velocity fields is limited by the size of interrogation windows and the boundary layer was not captured well. In this study, single-pixel ensemble correlation algorithm was applied to analyze flow near the surface of an axisymmetric boattail model. The initial images data was obtained by experimental methods with the setup of PIV measurement. The results showed that the new algorithm was considerably improved resolution of flow fields near the surface and could be used to measure boundary-layer profile. Detailed characteristics of boundary-layer profile at different flow conditions were discussed. Interestingly, boundary- layer profile does not change much before the shoulder. However, the size of separation bubble on the boattail surface highly decreases with increasing Reynolds number. The study provides initial results of flow fields, which could be useful for further investigation of drag reduction by numerical and experimental techniques. Keywords: Single-pixel ensemble correlation; PIV measurement; boattail model; boundary layer. 1. Introduction Reducing base drag and improving performance of the blunt-base vehicle is a big challenge for aerodynamic and fluid researchers in many years. Among of many devices for drag reductions such as base bleed, lock-vortex afterbody, splitter plate, base cavity and boattail model, the boattail model shows high effective [1]. A boattail model is determined as an additional contour shape added to blunt base model. In fact, the boattail model was widely applied for missiles and projectiles at high speed flow [2, 3]. However, flow behavior around the boattail model and its effect on drag reduction of model is not fully understood at low-speed conditions [3, 4]. Major studies of flow behavior around the boattail model at low speed were conducted by Mair [1, 3]; Buresti [5]; Mariotti et al. [6, 7] and Tran et al. [8-10]. The results indicated that the flow around boattail models at low speed shows many different * Email: thehungmfti@gmail.com Selected Papers of Young Researchers - 2020 90 features to that of high speed. Additionally, since flow around the base is very sensitive to disturbance at low speed condition, measurement the boundary-layer profile of boattail model is significantly complicated. Generally, it is a big challenge for both experimental technique and data processing. Consequently, improving measurement and data processing techniques are essential for further discussion of flow behavior and drag reduction strategy. Particle image velocimetry (PIV) measurement provides a potential technique in analyzing velocity fields [11]. In fact, PIV measurement is a non-instrusive measurement technique, which does not disturb the flow fields. The working principle of PIV measurement technique is to measure the displacement of small tracer particles over a short time interval. For data processing, cross-correlation algorithm is applied for small interrogation window in the first and second frames. The size of interrogation window often ranges from 8×8 pixels to 64×64 pixels, which reduces the resolution of the velocity fields. Additionally, since the interrogation windows could cover the wall region, the boundary-layer profile is not captured correctly. One way to improve the results is to zoom-in boundary region and to repeat experiments for different areas. Clearly, that process requires high effort and consumes a lot of time. The purpose of the current study is to apply a novel data processing technique for analyzing boundary layer of axisymmetric boattail model. In details, single-pixel ensemble correlation algorithm, which was proposed by Westerweel et al. [12], is applied to obtain high resolution of the flow fields near the wall. In fact, the algorithm was applied in previous studies for micro-PIV measurement and was validated by Kahler et al. [13]. However, the application for boundary-layer measurement of axisymmetric boattail was not illustrated. We will use the data of tranditional cross- correlation algorithm far from the wall to validate results of the current methods. This study shows that both algorithms provide good results for flow far from the wall. Additionally, the velocity profiles near the wall by single-pixel ensemble correlation are much improved by comparison to that of traditional cross-correlation algorithms. Consequently, PIV measurement with single-pixel ensemble correlation algorithm provides a promising tool to measure the boundary layer of moving object. The flow behavior around boattail model of 20º and its boundary-layer thickness at different Reynolds number will be discussed in detail in this study. Processing results could be used as initial data for further investigation of afterbody flow by both numerical and experimental methods. Journal of Science and Technique - N.208 (6-2020) - Le Quy Don Technical University 91 2. Experimental setup The experimental setup was similar to the one by Tran et al. [8, 10]. In the measurement, axisymmetric boattail model was supported in wind tunnel by a strut with cross section of NACA 0018. The diameter D of the model is 30 mm and the total length L is 251 mm. At the end the cylinder part, a conical boattail with fixed length of Lb = 0.7 D and angle of β = 20º was added (Fig. 1). Fig. 1. Model in wind tunnel test For PIV measurement, a laser was placed on the top to illuminate particles in the test section. Double-pulsed Nd-YLF Laser (LDY-303, 527 nm, Litron Lasers) was employed for the experiments. Laser sheet is setup at minimum thickness, which was around 1 mm. Time interval between double frame was varied by speed of wind tunnel in the range from 4 µs to 8 µs. The maximum movement of particles in images of a double frame was around 6 pixels. For generating luminescent particles inside test section, smoke generator LSG-500S was employed. The smoke generator has five laskin nozzles and can provide air with smoke particles of around 1 µm in diameter and 25 m3/h in volume. A high-speed camera Phantom V611 was placed on one side of test section to record particles movements around the model. The camera had a resolution of 1280×800 pixels and was equipped with a Nikon lens 100 mm f2.8. Additionally, an extension tube (36 mm) was also placed in front of the lens to increase magnification of the measurement section. The camera angle with dimensions of 40 mm × 25 mm was illustrated by red dashed line as shown in Fig. 2. The resolution of image reached around 32.5 pixels/mm. In addition, the camera was setup at 600 fps and movement of particles was recorded at around 9 s. Selected Papers of Young Researchers - 2020 92 Experiments were conducted at four different velocities from 22 m/s to 45 m/s, which gave the based-diameter Reynolds number from 4.34 × 104 to 8.89 × 104. a) Schematic of PIV measurement b) Wind tunnel mesurement Fig. 2. Setup of PIV measurement and wind tunnel test for flow velocity measurement 3. Measurement technique For data processing, the cross-correlation algorithm divides the first image into small interrogation areas (interrogation windows). After that, the cross-correlations of those windows with the second image are calculated. The position of maximum cross- correlation shows the displacement of the interrogation windows in the second images. Since the time interval between the first and second images were known and displacement of interrogation windows was calculated, the velocity fields can be obtained. The formula for cross-correlation is shown as: 1 2( ) ( ) ( ) W R s I X I X s dX  (1) where I1 and I2 present the first and second image, X is the coordinate, W is the size of interrogation window and s is the displacement. As the velocity of each interrogation windows is obtained, velocity fields of the whole image could be constructed. The method allows obtaining instantaneous velocity field from two images at different small time. By averaging instantaneous values at different time interval, the mean velocity fields can be found. Journal of Science and Technique - N.208 (6-2020) - Le Quy Don Technical University 93 The size of the interrogation window often ranges from 8×8 pixels to 64×64 pixels. Obviously, it reduces the resolution of velocity fields by comparison to image data. Additionally, it is very difficult to capture the flow fields near the wall, where the number of particles is significantly limited and the interrogation windows contain boundary of models and free air. To overcome the disadvantage of the cross-correlation algorithm, the single-pixel resolution ensemble correlation algorithm is used for data processing. The algorithm calculates cross-correlation coefficient for a single position of the first image and the interrogation windows in the second image from a group of double frames [12]. In more detail, information of each pixel in the first serial images and second serial images from a huge number of images was collected. Then, cross-correlation of each pixel in the first images with the second images was calculated. As the results, the displacement of each pixel in the first serial images can be found and velocity fields can be obtained. Clearly, by comparison to cross-correlation algorithm which uses spatial domain, the single- pixel resolution ensemble correlation uses temporal domain for calculating displacement of the particles. To obtain the highly accurate results, a large number of double frames is requested. Since single pixel is processed separately, the resolution of velocity fields is the same with the size of image. Additionally, flow near the wall is measured highly accurate. The principles of the cross-correlation algorithm and the single-pixel ensemble correlation are shown in Fig. 3. a) Cross-correlation algorithm b) Single-pixel ensemble correlation algorithm Fig. 3. Conventional and single-pixel ensemble correlation algorithm for data processing Selected Papers of Young Researchers - 2020 94 The formula of cross-correlation in single-pixel ensemble correlation algorithm is shown as: ( ) ( ) 1 2 1 1( ) ( ) ( ) N i i i R s I X I X s N    (2) where N is the total number of double image. In this study, 5400 double-frame images are processed to obtain the average velocity field. Since the maximum displacement of particles from first to second frames is around 6 pixels, the displacement of each pixel in the first images was searched in a surrounding window of 25×25 pixels in the second image frames to reduce calculated time. 4. Results and discussions 4.1. Comparison between cross-correlation and single-pixel algorithms Figure 4 presents streamwise velocity fields around the boattail model at Reynolds number of Re = 4.34 × 104. Here, the x-axis was normalized by boattail length while the z-axis was normalized by diameter of model. Both methods provide sufficiently good results far from the model. However, cross-correlation algorithm shows unclear results near the shoulder and around the edges of image. Clearly, cross-correlation algorithm shows some uncertain results near the borderlines, as it was discussed in Section 3. The results were improved largely by single-pixel ensemble correlation method, where clear velocity fields were illustrated. Consequently, the single-pixel ensemble correlation algorithm shows highly effective in determining flow behavior near the surface of model. a) Cross-correlation algorithm b) Single-pixel method Fig. 4. Velocity fields in two measurement methods at Re = 4.34×104 A comparison of the boundary-layer profile at x/D = -0.2 (6 mm before the shoulder) are shown in Fig. 5. The y-axis shows distance from the wall of the model. At 7 mm above the boattail surface, the velocity profile of two measurement methods is highly consistent. However, cross-correlation algorithm did not capture well the Journal of Science and Technique - N.208 (6-2020) - Le Quy Don Technical University 95 boundary layer near the wall. It can be explained that the interrogation window covers the wall region and processing results are affected. In the opposite site, the single-pixel ensemble correlation algorithm improved remarkably the velocity profile. Fig. 5. Boundary-layer profile from two algorithms 4.2. Mean velocity fields a) Re = 4.34 × 104 b) Re = 5.92 × 104 c) Re = 7.30 × 104 d) Re = 8.89 × 104 Fig. 6. Streamwise velocity fields on symmetric vertical plan at β = 20° Selected Papers of Young Researchers - 2020 96 The mean flow velocity in the vertical plane was shown in Fig. 6 for different flow conditions. The black dots show position of zero velocity streamline (dividing streamline). For all case, the flow is highly bent around the shoulder, which is affected by boattail geometry. A small separation bubble region is observed on the surface. Interestingly, the size of separation bubble decreases quickly with increasing Reynolds number from Re = 4.34 × 104 to Re = 8.89 × 104. At Reynolds number around Re = 8.89 × 104, separation bubble region becomes narrow and flow above the boattail is mainly affected by the geometry. It is expected that the separation bubble will be disappeared at higher Reynolds number or high Mach number conditions. The separation bubble flow is, therefore, a typical regime at low-speed conditions and was captured well by the single-pixel ensemble correlation algorithm. Note that previous study by Lavrukhin and Popovich [14] did not show a separation bubble for a wide range of Mach number conditions. 4.3. Characteristics of separation and reattachment on the boattail surface Fig. 7. Separation and reattachment positions on boattail surface at different Reynolds number conditions (S is separation position, R is reattachment position) Figure 7 shows separation and reattachment position on the boattail surface by PIV measurement and global luminescent oil film (GLOF) skin-friction measurement, which was obtained from previous study by Tran et al. [10]. The GLOF measurement captured skin-friction fields on the surface by a luminescent oil-film layer. The separation and reattachment positions by PIV measurement are determined by streamwise velocity along the boattail surface changing to negative and positive, respectively. The separation positions in both two methods show analogous results. At high Reynolds numbers, reattachment positions present similar results for two methods. However, at Reynolds number around Re = 4.34 × 104, results of both methods show Journal of Science and Technique - N.208 (6-2020) - Le Quy Don Technical University 97 remarkably different. It can be explained that the movement of air near reattachment position at low speed (Re = 4.34 × 104) is sufficient small and the number of particles near the boattail surface is not enough to obtain good data for PIV measurement processing. Additionally, due to unsteady behavior, the reattachment is often formed a large region on the surface. 4.4. Boundary-layer velocity profiles Figure 8 shows the boundary-layer profile for different Reynolds numbers tested at x/Lb = -0.2 (6 mm before the shoulder). The velocity profiles are averaged from 10 pixels surrounding measurement point in horizontal direction. Boundary-layer thickness δ is identified by a distance from wall surface to the position where streamwise velocity reaches to 95% free-stream velocity. The boundary-layer thickness is around δ = 2.8 mm and changes slightly for different flow conditions. Fig. 8. Boundary measurement at different Reynolds number As boundary-layer profiles are obtained, the displacement thickness δ*, momentum thickness θ and shape factor H can be calculated. Those parameters are shown by below equations: * * 0 0 ( ) ( ) ( )1 , 1 ,u z u z u zdz dz H U U U                            (3) The laminar boundary layer is characterized by the shape factor around H = 2.59 (Blasius boundary layer), while the turbulent boundary layer is characterized by H = 1.3-1.4. Selected Papers of Young Researchers - 2020 98 Table 1 shows boundary-layer parameters at Reynolds number of Re = 4.34 × 104. Clearly, boundary layer is fully turbulent before shoulder, which is shown by a shape factor of around H = 1.3. Tab. 1. Characteristics of boundary layer δ99/D δ*/D θ/D H 0.0933 0.0180 0.0134 1.34 Figure 9 shows boundary-layer profiles at different positions on the boattail surfaces for two cases of Reynolds numbers Re = 4.34 × 104 and Re = 8.89 × 104. The black dashed line presents dividing streamline at Re = 4.34 × 104. Clearly, the thickness of separation bubble at low Reynolds number is very high, which can be observed clearly from boundary-layer profile. However, separation bubble becomes smaller at high Reynolds number and it is not clearly illustrated. The figure also indicates that the thickness of boundary layer increases largely on the rear part of boattail model. Clearly, increasing thickness of boundary layer leads to a decreasing suction behind the base. Consequently, base drag of boattail model decreases. Fig. 9. Boundary profile at different positions on the boattail surface The relative thickness of boundary layer at different positions was shown in the Fig. 10 for two Reynolds number of Re = 4.34 × 104 and Re = 8.89 × 104. The different boundary-layer thickness at x/Lb = -0.2 is small, as it was indicated before. However, boundary-layer thickness changes quickly near the shoulder and in the boattail surface. As the Reynolds number increases, the separation bubble becomes smaller and the thickness of boundary layer near the shoulder is reduced. In fact, the changes of boundary-layer thickness occurred before the shoulder, which is caused by increasing streamwise velocity. However, at x/Lb > 0.2, the thickness of boundary layer increases Journal of Science and Technique - N.208 (6-2020) - Le Quy Don Technical University 99 with Reynolds number. Clearly, at high Reynolds number, the kinetic energy is remarkably lost on the boattail region and velocity recovery is lower. The high thickness of boundary layer near the base edge leads to a weaker near-wake and a decrease of base drag [15]. The results of boundary-layer profile also show some unsmooth changes near the base edge. It occurs from unperfected smooth of glass window, which uses to cover the test section of wind tunnel. To improve the results, further experiment should be conducted. However, this region is far from shoulder and does not affect our discussions. Fig. 10. Boundary-layer thickness 4.5. Skin-friction examination For turbulent flow in a smooth wall and non-pressure gradient, a log-law region exists above the buffer layer. In this region, the velocity changes as a logarithmic function of distance to wall surface [16]. The existence of the logarithmic law allows estimation of wall shear stress of the model. In more details, relation among those parameters is shown as: 1 lnu z C      (4) where uu u   , zu z     are non-dimensional velocity and distance from the wall and wu    is the friction vel